Analysis of Internet Topologies

Transcription

1 Analysis of Internet Topologies Ljiljana Trajković Communication Networks Laboratory School of Engineering Science Simon Fraser University, Vancouver, British Columbia Canada

2 Roadmap Internet topology and the datasets Spectrum of a graph and power-laws Power-laws and the Internet topology Spectral analysis of the Internet graph Conclusions and references CAS Society: DLP talk 2

3 Roadmap Internet topology and the datasets Spectrum of a graph and power-laws Power-laws and the Internet topology Spectral analysis of the Internet graph Conclusions and references CAS Society: DLP talk 3

4 lhr (535,102 nodes and 601,678 links) CAS Society: DLP talk 4

5 Internet graph Internet is a network of Autonomous Systems: groups of networks sharing the same routing policy identified with Autonomous System Numbers (ASN) Autonomous System Numbers: Internet topology on AS-level: the arrangement of ASes and their interconnections Analyzing the Internet topology and finding properties of associated graphs rely on mining data and capturing information about Autonomous Systems (ASes) CAS Society: DLP talk 5

6 Internet routing protocol Border Gateway Protocol (BGP): inter-as protocol used to exchange network reachability information among BGP systems reachability information is stored in routing tables CAS Society: DLP talk 6

7 Internet AS-level data Source of data are routing tables: Route Views: most participating ASes reside in North America RIPE (Réseaux IP européens): most participating ASes reside in Europe The BGP routing tables are collected from multiple geographically distributed BGP Cisco routers and Zebra servers. Analyzed datasets were collected at 00:00 am on July 31, 2003 and 00:00 am on July 31, CAS Society: DLP talk 7

8 Internet AS-level data Data used in prior research (partial list): Faloutsos, 1999 Chang, 2001 Vukadinovic, 2001 Gkantsidis, 2003 Route Views Yes Yes Yes Yes RIPE No Yes No Yes These research results have been used in developing Internet simulation tools: power-laws are employed to model and generate Internet topologies: BA model, BRITE, Inet CAS Society: DLP talk 8

9 Data sets Concerns about the use of the two datasets: different observations about AS degrees: power-law distribution: Route Views [Faloutsos, 1999] Weibull distribution: Route Views + RIPE [Chang, 2001] data completeness: RIPE dataset contains ~ 40% more AS connections and 2% more ASs than Route Views [Chang, 2001] CAS Society: DLP talk 9

10 Route Views and RIPE: statistics Route Views and RIPE samples collected on May 30, 2003 Number of AS paths Probed ASs AS pairs Route Views 6,398,912 15,418 34,878 RIPE 6,375,028 15,433 35,225 AS pair: a pair of connected ASs 15,369 probed ASs (99.7%) in both datasets are identical 29,477 AS pairs in Route Views (85%) and in RIPE (84%) are identical CAS Society: DLP talk 10

11 Degree distributions: 2003 data Consider all ASs with assigned AS numbers AS degree distribution in Route Views and RIPE datasets: CAS Society: DLP talk 11

12 Route Views RIPE Core ASs ASs with largest degrees AS Degree AS Degree CAS Society: DLP talk 12

13 Route Views RIPE AS Degree AS Degree Core ASs ASs with largest degrees 16 of the core ASs in Route Views and RIPE are identical Core ASs in Route Views have larger degrees than core ASs in RIPE CAS Society: DLP talk 13

14 Internet topology Datasets are collected from Border Gateway Protocols (BGP) routing tables. The Internet topology is characterized by the presence of various power-laws observed when considering: node degree vs. node rank node degree frequency vs. degree number of nodes within a number of hops vs. number of hops eigenvalues of the adjacency matrix and the normalized Laplacian matrix vs. the order of the eigenvalues Faloutsos et al., 1999 and Siganos et al., CAS Society: DLP talk 14

15 Roadmap Internet topology and the datasets Power-laws and spectrum of a graph Power-laws and the Internet topology Spectral analysis of the Internet graph Conclusions and references CAS Society: DLP talk 15

16 Spectrum of a graph Normalized Laplacian matrix NL(G): NL( i, 1 j) = 0 1 d d i j 0 d i and d j are degrees of node i and j, respectively The spectrum of NL(G) is the collection of all eigenvalues and contains 0 for every connected graph component. if if i = i and j and d otherwise i j are adjacent Chung et al., CAS Society: DLP talk 16

17 Power laws: node degree vs. rank The graph nodes v are sorted in decreasing order based on their degrees d v and are indexed with a sequence of numbers indicating their ranks r v. The (r v, d v ) pairs are plotted on the log-log scale. The power-law implies: d R v r v where v is the node number and R is the node degree powerlaw exponent., CAS Society: DLP talk 17

18 Power laws: CCDF of a node degree The frequency of a node degree is equal to the number of nodes having the same degree. The complementary cumulative distribution function (CCDF) D d of a node degree d is equal to the number of nodes having degree less than or equal to d, divided by the number of nodes. The power-law implies: D, D where D is the CCDF power-law exponent. d d CAS Society: DLP talk 18

19 Power laws: eigenvalues The eigenvalues λ i of the adjacency matrix and the normalized Laplacian matrix are sorted in decreasing order and plotted versus the associated increasing sequence of numbers i representing the order of the eigenvalue. The power-law for the adjacency matrix implies: λ ai i ε, The power-law for the normalized Laplacian matrix implies: λ Li where ε and L are the eigenvalue power-law exponents. i L, CAS Society: DLP talk 19

20 Analysis of datasets Calculated and plotted on a log-log scale are: node degree vs. node rank frequency of node degree vs. node degree eigenvalues vs. index Linear regression is used to determine the correlation coefficient between the regression line and the plotted data. A high correlation coefficient between the regression line and the plotted data indicates the existence of a power-law, which implies that node degree, frequency of node degree, and eigenvalues follow a power-law dependency on the rank, node degree, and index, respectively CAS Society: DLP talk 20

21 Analysis of datasets The power-law exponents are calculated from the linear regression lines 10 a x b, with segment a and slope b when plotted on a log-log scale. Source of data are routing tables: Route Views: RIPE (Réseaux IP européens): Faloutsos et al., 1999 and Chen et al., CAS Society: DLP talk 21

22 Roadmap Internet topology and the datasets Spectrum of a graph and power-laws Power-laws and the Internet topology Spectral analysis of the Internet graph Conclusions and references CAS Society: DLP talk 22

23 Route Views 2003 dataset The node degree power-law exponent R = The correlation coefficient = CAS Society: DLP talk 23

24 Route Views 2008 dataset The node degree power-law exponent R = The correlation coefficient = CAS Society: DLP talk 24

25 Confidence intervals: node degree CAS Society: DLP talk 25

26 Confidence intervals Six samples were randomly selected from 2003 and 2008 Route Views and RIPE datasets Each dataset is smaller than 30, with unknown standard deviation T-distribution was used to predict the confidence interval at 95% confidence level X t / 2( s / n ) < μ < X tx / 2( s / n ) x + t x / X : the sample mean 2 : the t-distribution s : sample standard deviation n : number of samples μ : population mean CAS Society: DLP talk 26

27 Route Views 2003 dataset CCDF power-law exponent D = The correlation coefficient = CAS Society: DLP talk 27

28 Route Views 2008 dataset CCDF power-law exponents D = The correlation coefficients = CAS Society: DLP talk 28

29 Confidence intervals: CCDF CAS Society: DLP talk 29

30 Route Views 2003 dataset The eigenvalue power-law exponents ε = The correlation coefficient = CAS Society: DLP talk 30

31 Route Views 2008 dataset The eigenvalue power-law exponent ε = The correlation coefficient = CAS Society: DLP talk 31

32 Confidence intervals: adjacency matrix CAS Society: DLP talk 32

33 Route Views 2003 dataset The eigenvalue power-law exponent L = The correlation coefficient = CAS Society: DLP talk 33

34 Route Views 2008 dataset The eigenvalue power-law exponent L = The correlation coefficient = CAS Society: DLP talk 34

35 Confidence intervals: normalized Laplacian matrix CAS Society: DLP talk 35

36 Roadmap Internet topology and the datasets Spectrum of a graph and power-laws Power-laws and the Internet topology Spectral analysis of the Internet graph Conclusions and references CAS Society: DLP talk 36

37 Spectral analysis of Internet graphs We calculate the second smallest and the largest eigenvalues and associated eigenvectors of normalized Laplacian matrix. Each element of an eigenvector is associated with the AS having the same index. ASes are sorted in the ascending order based on the eigenvector values and the sorted AS vector is then indexed. The connectivity status is equal to one if the AS is connected to another AS or zero if the AS is isolated or is absent from the routing table CAS Society: DLP talk 37

38 Spectral analysis of Internet graphs The second smallest eigenvalue, called "algebraic connectivity" of a normalized Laplacian matrix, is related to the connectivity characteristic of the graph. Elements of the eigenvector corresponding to the largest eigenvalue of the normalized Laplacian matrix tend to be positioned close to each other if they correspond to AS nodes with similar connectivity patterns constituting clusters. Gkantsidis et al., CAS Society: DLP talk 38

39 Characteristic valuation: example The second smallest eigenvector: 0.1, 0.3, -0.2, 0 AS1(0.1), AS2(0.3), AS3(-0.2), AS4(0) Sort ASs by element value: AS3, AS4, AS1, AS2 AS3 and AS1 are connected connectivity status 1 0 AS3 AS4 AS1 AS2 index of elements CAS Society: DLP talk 39

40 Spectral analysis: observations The second smallest eigenvector: separates connected ASs from disconnected ASs Route Views and RIPE datasets are similar on a coarser scale The largest eigenvector: reveals highly connected clusters Route Views and RIPE datasets differ on a finer scale CAS Society: DLP talk 40

41 Route Views 2003 dataset Spectral views of the AS connectivity based on the second smallest eigenvalue CAS Society: DLP talk 41

42 Route Views 2008 dataset Spectral views of the AS connectivity based on the second smallest eigenvalue CAS Society: DLP talk 42

43 Route Views 2003 dataset Spectral views of the AS connectivity based on the largest eigenvalue CAS Society: DLP talk 43

44 Route Views 2008 dataset Spectral views of the AS connectivity based on the largest eigenvalue CAS Society: DLP talk 44

45 Roadmap Internet topology and the datasets Spectrum of a graph and power-laws Power-laws and the Internet topology Spectral analysis of the Internet graph Conclusions and references CAS Society: DLP talk 45

46 Conclusions We have evaluated collected data from the Route Views and RIPE projects and have confirmed the presence of power-laws in graphs capturing the AS-level Internet topology. Spectral analysis techniques revealed distinct clustering characteristics of Route Views and RIPE datasets The analysis also captured historical trends in the development of the Internet topology over the past five years. In spite of the Internet growth, increasing number of users, and the deployment of new network elements, powerlaw exponents have not changed substantially CAS Society: DLP talk 46

47 Conclusions These power-law exponents do not capture every property of a graph and are only one measure used to characterize the Internet. However, spectral analysis based on the normalized Laplacian matrix indicated visible changes in the clustering of AS nodes and the AS connectivity CAS Society: DLP talk 47

48 References M. Najiminaini, L. Subedi, and Lj. Trajkovic, "Analysis of Internet topologies: a historical view," presented at IEEE International Symposium Circuits and Systems, Taipei, Taiwan, May J. Chen and Lj. Trajkovic, "Analysis of Internet topology data," Proc. IEEE International Symposium on Circuits and Systems, Vancouver, BC, Canada, May 2004, vol. IV, pp M. Faloutsos, P. Faloutsos, and C. Faloutsos, On power-law relationships of the Internet topology, Proc. ACM SIGCOMM, Computer Communication Review, vol. 29, no. 4, pp , Sept G. Siganos, M. Faloutsos, P. Faloutsos, and C. Faloutsos, "Power-laws and the AS-level Internet topology," IEEE/ACM Trans. Networking, vol. 11, no. 4, pp , Aug A. Medina, I. Matta, and J. Byers, "On the origin of power laws in Internet topologies," Proc. ACM SIGCOMM 2000, Computer Communication Review, vol. 30, no. 2, pp , Apr CAS Society: DLP talk 48

49 References L. Gao, "On inferring autonomous system relationships in the Internet," IEEE/ACM Trans. Networking, vol. 9, no. 6, pp , Dec D. Vukadinovic, P. Huang, and T. Erlebach, On the Spectrum and Structure of Internet Topology Graphs, in H. Unger et al., editors, Innovative Internet Computing Systems, LNCS2346, pp Springer, Berlin, Germany, Q. Chen, H. Chang, R. Govindan, S. Jamin, S. Shenker, and W. Willinger, "The origin of power laws in Internet topologies revisited," Proc. INFOCOM, New York, NY, USA, Apr. 2002, pp H. Chang, R. Govindan, S. Jamin, S. Shenker, and W. Willinger, "Towards capturing representative AS-level Internet topologies," Proc. of ACM SIGMETRICS 2002, New York, NY, June 2002, pp H. Tangmunarunkit, R. Govindan, S. Jamin, S. Shenker, and W. Willinger, "Network topology generators: degree-based vs. structural," Proc. ACM SIGCOMM, Computer Communication Review, vol. 32, no. 4, pp , Oct CAS Society: DLP talk 49

50 References C. Gkantsidis, M. Mihail, and E. Zegura, "Spectral analysis of Internet topologies, Proc. of Infocom 2003, San Francisco, CA, Mar. 2003, vol. 1, pp S. Jaiswal, A. Rosenberg, and D. Towsley, "Comparing the structure of power-law graphs and the Internet AS graph," Proc. 12th IEEE International Conference on Network Protocols, Washington DC, Aug. 2004, pp F. R. K. Chung, Spectral Graph Theory. Providence, Rhode Island: Conference Board of the Mathematical Sciences, 1997, pp M. Fiedler, "Algebraic connectivity of graphs," Czech. Math. J., vol. 23, no. 2, pp , CAS Society: DLP talk 50

51 lhr (535,102 nodes and 601,678 links) CAS Society: DLP talk 51

52 lhr (535,102 nodes and 601,678 links) CAS Society: DLP talk 52

53 lhr (535,102 nodes and 601,678 links) CAS Society: DLP talk 53

54 Round-trip time measurements (63,631 nodes and 63,630 links) CAS Society: DLP talk 54

55 Round-trip time measurements (63,631 nodes and 63,630 links) CAS Society: DLP talk 55

56 Resources CAIDA: The Cooperative Association for Internet Data Analysis Walrus - Gallery: Visualization & Navigation Walrus - Gallery: Abstract Art CAS Society: DLP talk 56